p–ISSN: 2723 - 6609 e-ISSN: 2745-5254
Vol. 5, No. 7 July 2024 http://jist.publikasiindonesia.id/
Jurnal Indonesia Sosial Teknologi, Vol. 5, No. 7, July 2024 3412
Design of Forecasting Electrical Power of Ultra-Short-Term
Solar Power Using the Hybrid Model K-Nearest Neighbors
LSTM
Tri Wahyu Yulianto
1*
, Unit Three Kartini
2
, Bambang Suprianto
3
Universitas Negeri Surabaya, Indonesia
Email:
1*
2
,
3
*Correspondence
ABSTRACT
Keywords: solar PV, solar
radiation, PV K-NN panel
temperature, LSTM.
For the application of renewable energy at the airport, the
use of solar power requires certainty of the electricity
produced. The certainty of electricity generated from solar
power can be predicted using machine learning methods.
Predictions made on PV electrical power output are based on
historical data from direct measurements of solar PV
parameters, including solar radiation and PV panel
temperature. Various types of machine learning methods for
predicting PV output power have been used in previous
studies with different evaluation values of prediction results.
In this study, the author conducted a hybrid K-NN method
with LSTM to predict the PV electrical power of solar PV
output with solar radiation parameters and PV panel
temperature. After making predictions using this method,
excellent RSME results were obtained with a value of
0.015424830635781967. The results of the PV output power
value graph in this prediction are also very good, where the
predicted value is close to the value of the testing data or
actual data.
Introduction
For the application of new and renewable energy at the airport, the use of solar
power is felt to be the most appropriate because, in addition to being easy to install, it can
be installed on the rooftop of the building or the ground, as well as the ease of availability
of solar power plants on the market (PT Angkasa Pura 2, 2022a).
Solar PV is a type of power plant with a working system that converts solar energy
into electrical energy with several variables that affect the production of electrical energy
or optimal PV power output (Rifa’i, Ananda, & Fadhli, 2018). The first variable that
affects PV power output is PV module temperature and ambient temperature, two types
of temperature that affect PV power output. The temperature of the PV module has a
stronger influence of about 20~30% of the ambient temperature for the output of the PV
power produced (Institute of Electrical and Electronics Engineers, n.d.). The second
variable that influences PV power output is solar irradiance. In the tests that have been
Design of Forecasting Electrical Power of Ultra-Short-Term Solar Power Using the Hybrid
Model K-Nearest Neighbors LSTM
Jurnal Indonesia Sosial Teknologi, Vol. 5, No. 7, July 2024 3413
carried out, the distribution of solar radiation (solar irradiance) and the temperature of the
PV module influence the power output performance produced by the PV module
(Nurlindah, Mustami, & Musdalifah, 2020).
Various studies in the form of testing and prediction experiments to simulate the
performance of PV modules based on historical data have been carried out with the help
of mathematical calculations, this is because 2 factors that affect the performance of PV
modules to produce power output have different conditions at different times and
conditions (Liu et al., 2021). One of the measurements to determine/assess the prediction
results of a method is by using RSME (Root Square Mean Error). Prediction experiments
have been carried out using Times series and ANN modelling techniques with the KNN
prediction model producing a lower RSME value if in the process of predicting the "k"
value of the K-NN method is getting larger (Sivakumar et al., 2022). Experiments by
combining several methods (hybrid) have also been carried out, namely based on
correlation data between PV module power output, solar irradiance and PV module
temperature using ARIMA, ANFIS, ANN, and SVM modelling techniques in the first
stage of modelling, where the results in the first stage are then combined with the
prediction method using GA (Genetic Algorithm) which produces better prediction data
(Ozbek, Yildirim, & Bilgili, 2022).
In this study, the author uses a combination of methods to predict the power output
of PV modules with the K-Nearest Network (KNN) modelling method and the KNN
prediction data is used as new data to be modelled and re-predicted using the Long short-
term memory (LSTM) method (Wu, Chen, & Abdul Rahman, 2014). To obtain better
prediction results, where later the prediction results will be measured/reviewed by the
RSME method, the author selects the prediction period on the daily data in the morning
and afternoon when the PV module produces power output only (Purwantoro, Kartini,
Suprianto, & Agung, 2022).
Solar irradiance and PV module temperature are very dynamic variables and their
changes are affected by the surrounding environmental conditions for each period, so the
value is not a time series (Asfah & Kartini, 2020). The RSME value does influence
determining the accuracy value of a prediction method used to predict the module's PV
power output every period. However, the selection of the prediction method and the
period to be carried out also have a role and influence on the predictions produced (Agam
& Kartini, 2020).
The novelty of previous research/Novelty of the research conducted concerning
previous research is to make a very short-term forecast design at solar power plants
located at Soekarno-Hatta airport based on meteorological data, namely temperature and
solar irradiance using the K-Nearest Neighbors hybrid method and the Long Short Term
Memory (KNN-LSTM) method.
Research Methods
The object of this research is the Soekarno-Hatta Airport Solar Power Plant installed
in Terminal 2E. Solar Power Plant in Terminal 2E can capture solar energy with a
Tri Wahyu Yulianto, Unit Three Kartini, Bambang Suprianto
Jurnal Indonesia Sosial Teknologi, Vol. 5, No. 7, July 2024 3414
capacity of 636.65kWp and the minimum amount of electrical energy that can be
generated is 1887MWh. This solar power plant has 1,168 PV modules and the capacity
of each PV module is 545Wp.
Data collection
The data collection time in this study is September 27 and 29, 2023. The data used
in this study are the following data:
Time
This data is the time when other data were taken. This data shows the time in hours
and minutes. The period in 1 day of research data was taken from 05.30 to 17.30 with a
time range of 5 minutes for each data.
PV Output
This data is the value of electrical power generated from solar PV, where this data
value is obtained according to the time in the "Time" data. The unit amount of electrical
power generated from solar power is Wh (Watt hour).
Solar Irradiance
This data is the solar radiation value contained regarding the PV module of the solar
PV, where this data value is obtained according to the time in the "Time" data. The unit
size of solar radiation that hits the PV module of the solar PV is W/m².
PV Temp
This data is the temperature value on the PV panel of the solar module, where this
data value is obtained according to the time in the "Time" data. The unit size of PV panel
modules from solar PV is °C.
The data used in this study were collected using direct observation from the web
display to monitor and measure each parameter of the solar power plant.
Exploration Data Analysis (EDA)
Normalization is one of the most frequently used data preparation techniques. In
machine learning and data mining [6], this process helps to change the numeric column
values in the dataset to use a common scale. One of the challenges that exists in databases
is the existence of attributes with different units, ranges, and scales.
Applying data mining or machine learning algorithms to data with drastic ranges
can provide less accurate results. Data normalization is a basic element of data mining to
ensure that records in the dataset remain consistent.
In the normalization process, it is necessary to transform the data or convert the
original data into a format that allows efficient data processing. The main goal of data
normalization is to eliminate data redundancy (repetition) and standardize information for
better data workflows.
Data normalization is used to scale the data of an attribute so that it is within a
smaller range, such as -1 to 1 or 0 to 1. It is generally useful for classification algorithms.
The min-max normalization method converts a dataset into a scale ranging from 0 (min)
to 1 (max). The original data underwent linear modifications in this data normalization
procedure. The minimum and maximum values of the data are retrieved, and each value
is changed using the below formula:
Design of Forecasting Electrical Power of Ultra-Short-Term Solar Power Using the Hybrid
Model K-Nearest Neighbors LSTM
Jurnal Indonesia Sosial Teknologi, Vol. 5, No. 7, July 2024 3415
Xnorm = (x^'-min(x))/(max(x)-min(x)) (Newman(x) – Newman(x)) +
Newman(x)
x = date attribute
min(x) = Minimum absolute value of x
max(x) = Maximum absolute value of x
x' = the old value of each entry in the data
Newsmax(x) = Max value of x
Newman(x) = min value of x
Prediksi K-Nearest Neighbor
The KNN algorithm is a classification technique that determines the class of new
data by taking several K of the nearest data as the basis for comparison. This algorithm
works by comparing similarities or similarities between data (Ismail, 2018). The working
principle of K-Nearest Neighbor (KNN) is to find the closest distance between the data
to be evaluated and the nearest K-Nearest Neighbor in ththe e training data. The following
is the formula for finding distance using the Euclidian fms (Agusta, 200).
ℎ𝑎𝑙√∑ (𝑥
2𝑖
− 𝑥
1𝑖
)
2
𝑝
𝑖=1
Information:
p = Data dimension
i = Variable Data
x1 = Sample data
x2 = Test data or testing data
d = Distance
K-Nn Lstm Hybrid Method Prediction
The hybrid methods of KNN and LSTM combine the power of both algorithms to
improve prediction performance in some cases. KNN (K-Nearest Neighbors) is an
instance-based learning algorithm used primarily for classification and regression
problems, while LSTM (Long Short-Term Memory) is a type of recursive neural network
(RNN) architecture that is effective in modelling sequences and patterns in Time series
data. The process of the research implementation flow is illustrated in the flowchart in
Figure 1.
Tri Wahyu Yulianto, Unit Three Kartini, Bambang Suprianto
Jurnal Indonesia Sosial Teknologi, Vol. 5, No. 7, July 2024 3416
Begin
Data Collection
Setting Up Data Testing and Data
Training
Pre Processing Data
Modeling with KNN
Fittings with KNN
Predictions with KNN
Evaluation of KNN Prediction
Results (RSME Calculation)
RSME < 0.2
Set Data Testing and Data Training
Post KNN
Modeling with LSTM
Fittings with LSTM
Prediction with LSTM
-
Evaluation of ESG Prediction
Results (RSME Calculation)
Data Denormalization
Visualization of Results :
Predictions with KNN
Prediction with LSTM
RSME with KNN
RSME with LSTM
Finish
yes
No
Reset value
"k"
Design of Forecasting Electrical Power of Ultra-Short-Term Solar Power Using the Hybrid
Model K-Nearest Neighbors LSTM
Jurnal Indonesia Sosial Teknologi, Vol. 5, No. 7, July 2024 3417
Results and Discussion
In the flow process of research, the first step is to collect research data. The data
used in this study were collected using direct observation from the web display to monitor
and measure each parameter in the solar power plant. The following are the research data
that have been collected and preprocessed data.
Figure 2 Solar PV data on September 27, 2023,
Source: Personal, 2023
Figure 3
Solar PV Data Type Info on September 27, 2023,
Source: Personal, 2024
Figure 4
Description of Solar PV Data on September 27, 2023,
Source: Personal, 2024
Tri Wahyu Yulianto, Unit Three Kartini, Bambang Suprianto
Jurnal Indonesia Sosial Teknologi, Vol. 5, No. 7, July 2024 3418
Figure 5 Preprocessing of Solar PV Data on September 27, 2023,
Source: Personal, 2024
In the data preprocessing process, it is known that the number of data is 145 data
with the number of data columns as many as 4. After the data preprocessing process, a
training data set with a percentage of 0.7 of the total data and a testing data set with a
percentage of 0.3 of the total data were carried out. The results of the training and testing
data sets are shown in Figure 6.
Figure 6.
Testing Data Set and Training Data on Solar Power Plant Data on September 27, 2023,
Source: Personal, 2024
The next process is modelling, fitting and prediction using K-NN. This process is
carried out on several "k" values in K-NN with values from k=2 to k=10. This is done to
obtain the prediction value with the best evaluation results. After the prediction results
are obtained and the prediction results are evaluated, the prediction results for each "k"
are obtained as follows in the bar diagram Figure 7.
Design of Forecasting Electrical Power of Ultra-Short-Term Solar Power Using the Hybrid
Model K-Nearest Neighbors LSTM
Jurnal Indonesia Sosial Teknologi, Vol. 5, No. 7, July 2024 3419
Figure 7. RSME Prediction Score Using K-NN,
Source: Personal, 2024
According to Figure 10, the K-NN prediction with the best RSME value is in the
K-NN prediction with a value of "k" is 10 with a value of 0.11344051656173618.
Therefore, the results of the K-NN prediction with a value of k=10 are then used as initial
data to be predicted again using the LSTM method. The prediction results with LSTM
obtained better RSME results of 0.015424830635781967 or closer to "0" when compared
to the predicted RSME value which only used the K-NN k=10 method.
Meanwhile, the results of the prediction of the output PV power value based on
solar radiation parameters (Solar Irradiance) and PV panel temperature (PV Temp.) using
the two predictions above can be seen in Figure 8 and Figure 9.
Figure 8
Graph comparison of the values of each PV output power prediction method on the testing
data,
Source: Personal, 2024
Tri Wahyu Yulianto, Unit Three Kartini, Bambang Suprianto
Jurnal Indonesia Sosial Teknologi, Vol. 5, No. 7, July 2024 3420
Figure 9. Graph of PV Power Output Power Value Training Data and Values of Each PV
Output Power Prediction Method on Data Testing,
Source: Personal, 2024
Figure 10
RSME Prediction Score Using K-NN k=10 and K-NN Hybrid Prediction k=10 LSTM,
Source: Personal, 2024
Conclusion
Modelling and prediction of PV power output of solar PV at Soekarno-Hatta
Airport Terminal 2E have been carried out using solar radiation parameter data and PV
panel module temperature. The prediction method used is the K-NN method with a "k"
value of 10 and the RSME value of this prediction method is 0.11344051656173618. The
results of this prediction are used as preliminary data to re-predict the PV Output power
using the LSTM method because although the previous method has a good RSME value,
the PV Output power value has not approached the testing data as shown in Figure 11 and
Figure 12. After making predictions using the LSTM method, better RSME results were
obtained with a value of 0.015424830635781967. The results of the PV Output power
value graph display in this prediction are also better, where the predicted value is close to
the value of the testing data or actual data. So the first conclusion can be obtained that the
prediction of PV electrical power Output with solar radiation parameters and PV panel
temperature using the KNN prediction method with a value of k=10 obtained good
Design of Forecasting Electrical Power of Ultra-Short-Term Solar Power Using the Hybrid
Model K-Nearest Neighbors LSTM
Jurnal Indonesia Sosial Teknologi, Vol. 5, No. 7, July 2024 3421
prediction results. The following conclusion is that better prediction results can be
obtained when the prediction in the initial method is carried out hybrid method using the
LSTM method.
Tri Wahyu Yulianto, Unit Three Kartini, Bambang Suprianto
Jurnal Indonesia Sosial Teknologi, Vol. 5, No. 7, July 2024 3422
Bibliography
Agam, Masviki, & Kartini, Unit Three. (2020). Peramalan Daya Listrik PLTS On Grid
pada Rumah Tinggal Menggunakan Metode K-Nearest Neighbor Decomposition
Feed Forward Neural Network Berdasarkan Data Meteorologi. Jurnal Teknik
Elektro, 9(2).
Agusta, Yudi. (2007). K-means–penerapan, permasalahan dan metode terkait. Jurnal
Sistem Dan Informatika, 3(1), 47–60.
Asfah, Rani Fajriyah Islamiyati, & Kartini, Unit Three. (2020). Peramalan Radiasi Global
Matahari Jangka Pendek Menggunakan Modeltriple Exponential Smoothing-Feed
Forward Neural Network. Jurnal Teknik Elektro, 9(3), 677–684.
Ismail, Asep Maulana. (2018). Cara Kerja Algoritma k-Nearest Neighbor (k-NN).
Dipetik, 9(17), 2019.
Liu, Zhi Feng, Luo, Shi Fan, Tseng, Ming Lang, Liu, Han Min, Li, Lingling, & Mashud,
Abu Hashan Md. (2021). Short-term photovoltaic power prediction on modal
reconstruction: A novel hybrid model approach. Sustainable Energy Technologies
and Assessments, 45, 101048.
Nurlindah, Nurlindah, Mustami, Muh Khalifah, & Musdalifah, Musdalifah. (2020).
Manajemen Pendidik Dan Tenaga Kependidikan Dalam Meningkatkan Mutu
Pendidikan. Idaarah, 4(1), 40–51. https://doi.org/10.24252/idaarah.v4i1.13893
Ozbek, Arif, Yildirim, Alper, & Bilgili, Mehmet. (2022). Deep learning approach for one-
hour ahead forecasting of energy production in a solar-PV plant. Energy Sources,
Part A: Recovery, Utilization, and Environmental Effects, 44(4), 10465–10480.
Purwantoro, Kukuh Eko, Kartini, Unit Three, Suprianto, Bambang Suprianto, & Agung,
Achmad Imam. (2022). Prediksi Daya Listrik Jangka Sangat Pendek Pembangkit
Photovoltaic Berbasis Internet of Things Menggunakan Feed Forward Neural
Network. JURNAL TEKNIK ELEKTRO, 11(3), 386–396.
Rifa’i, Muhammad, Ananda, Rusydi, & Fadhli, Muhammad. (2018). Manajemen peserta
didik (Pengelolaan peserta didik untuk efektivitas pembelajaran).
Sivakumar, S., Neeraja, B., Jamuna Rani, M., Anandaram, Harishchander, Ramya, S.,
Padhan, Girish, & Gurusamy, Saravanakumar. (2022). Machine Learning Approach
on Time Series for PV‐Solar Energy. Advances in Materials Science and
Engineering, 2022(1), 6458377. https://doi.org/10.1155/2022/6458377
Wu, Yuan Kang, Chen, Chao Rong, & Abdul Rahman, Hasimah. (2014). A novel hybrid
model for short‐term forecasting in PV power generation. International Journal of
Photoenergy, 2014(1), 569249.
